More generic tuning updates

[Jan Hubicka <hubicka at ucw dot cz>]

Hi,
this patch enables X86_TUNE_SSE_TYPELESS_STORES for generic. This is because
it is currently enabled for both Bulldozer and Cores and I think in fact all
modern chips honnors it.  I am not sure why it is off for Atom/Slm and Bobcat.

Second change is to enable X86_TUNE_SSE_LOAD0_BY_PXOR for Bulldozer, Bobcat and
Generic.  This flag is already on for Intel chips as recomended by the optimization
manual.  Amdfam15 manual has similar recommendation but only for avx variant.
I believe it applies for pxor too and I verified it does not cause any regressions.

I am now benchmarking the next two flags I dropped FIXME on - the multiply by
immediate that are really specific for K8-based chips and I think it got
enabled elsewhere by a mistake.

Honza

Bootstrapped/regtested x86_64-linux, will commit it shortly.

	* config/i386/x86-tune.def: Enable X86_TUNE_SSE_TYPELESS_STORES
	for generic, enable X86_TUNE_SSE_LOAD0_BY_PXOR for Bulldozer,
	Bobcat and generic.

	* gcc.target/i386/avx256-unaligned-store-3.c: Update template for
	tuning change.
	* gcc.target/i386/avx256-unaligned-store-1.c: Likewise.
	* gcc.target/i386/pr49168-1.c: Likewise.
	* gcc.target/i386/pr49002-2.c: Likewise.
Index: testsuite/gcc.target/i386/avx256-unaligned-store-3.c
===================================================================
--- testsuite/gcc.target/i386/avx256-unaligned-store-3.c	(revision 203380)
+++ testsuite/gcc.target/i386/avx256-unaligned-store-3.c	(working copy)
@@ -18,5 +18,5 @@ avx_test (void)
 }
 
 /* { dg-final { scan-assembler-not "avx_storeupd256" } } */
-/* { dg-final { scan-assembler "vmovupd.*\\*movv2df_internal/3" } } */
+/* { dg-final { scan-assembler "vmovups.*\\*movv2df_internal/3" } } */
 /* { dg-final { scan-assembler "vextractf128" } } */
Index: testsuite/gcc.target/i386/pr49168-1.c
===================================================================
--- testsuite/gcc.target/i386/pr49168-1.c	(revision 203380)
+++ testsuite/gcc.target/i386/pr49168-1.c	(working copy)
@@ -2,7 +2,8 @@
 /* { dg-do compile } */
 /* { dg-options "-O2 -msse2 -mtune=generic" } */
 /* { dg-final { scan-assembler-not "movdqa\[\t \]*%xmm\[0-9\]\+,\[^,\]*" } } */
-/* { dg-final { scan-assembler "movdqu\[\t \]*%xmm\[0-9\]\+,\[^,\]*" } } */
+/* { dg-final { scan-assembler-not "movaps\[\t \]*%xmm\[0-9\]\+,\[^,\]*" } } */
+/* { dg-final { scan-assembler "movups\[\t \]*%xmm\[0-9\]\+,\[^,\]*" } } */
 
 void
 flt128_va (void *mem, __float128 d)
Index: testsuite/gcc.target/i386/pr49002-2.c
===================================================================
--- testsuite/gcc.target/i386/pr49002-2.c	(revision 203380)
+++ testsuite/gcc.target/i386/pr49002-2.c	(working copy)
@@ -11,4 +11,5 @@ void foo(const __m128d from, __m256d *to
 
 /* Ensure we store ymm, not xmm.  */
 /* { dg-final { scan-assembler-not "vmovapd\[\t \]*%xmm\[0-9\]\+,\[^,\]*" } } */
-/* { dg-final { scan-assembler "vmovapd\[\t \]*%ymm\[0-9\]\+,\[^,\]*" } } */
+/* { dg-final { scan-assembler-not "vmovaps\[\t \]*%xmm\[0-9\]\+,\[^,\]*" } } */
+/* { dg-final { scan-assembler "vmovaps\[\t \]*%ymm\[0-9\]\+,\[^,\]*" } } */
Index: testsuite/gcc.target/i386/avx256-unaligned-store-2.c
===================================================================
--- testsuite/gcc.target/i386/avx256-unaligned-store-2.c	(revision 203380)
+++ testsuite/gcc.target/i386/avx256-unaligned-store-2.c	(working copy)
@@ -24,5 +24,5 @@ avx_test (void)
 }
 
 /* { dg-final { scan-assembler-not "avx_storedqu256" } } */
-/* { dg-final { scan-assembler "vmovdqu.*\\*movv16qi_internal/3" } } */
+/* { dg-final { scan-assembler "vmovups.*\\*movv16qi_internal/3" } } */
 /* { dg-final { scan-assembler "vextract.128" } } */
Index: config/i386/x86-tune.def
===================================================================
--- config/i386/x86-tune.def	(revision 203380)
+++ config/i386/x86-tune.def	(working copy)
@@ -221,16 +221,14 @@ DEF_TUNE (X86_TUNE_SSE_PACKED_SINGLE_INS
    upper part undefined.  */
 DEF_TUNE (X86_TUNE_SSE_SPLIT_REGS, "sse_split_regs", m_ATHLON_K8)
 
-/* X86_TUNE_SSE_TYPELESS_STORES: Always movaps/movups for 128bit stores. 
-   FIXME: Shall we enable it for generic?  */
+/* X86_TUNE_SSE_TYPELESS_STORES: Always movaps/movups for 128bit stores.   */
 DEF_TUNE (X86_TUNE_SSE_TYPELESS_STORES, "sse_typeless_stores",
-	  m_AMD_MULTIPLE | m_CORE_ALL)
+	  m_AMD_MULTIPLE | m_CORE_ALL | m_GENERIC)
 
 /* X86_TUNE_SSE_LOAD0_BY_PXOR: Always use pxor to load0 as opposed to
-   xorps/xorpd and other variants.
-   FIXME: Shall we enable it buldozers and for generic?  */
+   xorps/xorpd and other variants.  */
 DEF_TUNE (X86_TUNE_SSE_LOAD0_BY_PXOR, "sse_load0_by_pxor",
-	  m_PPRO | m_P4_NOCONA | m_CORE_ALL)
+	  m_PPRO | m_P4_NOCONA | m_CORE_ALL | m_BDVER | m_BTVER | m_GENERIC)
 
 /* X86_TUNE_MEMORY_MISMATCH_STALL: Avoid partial stores that are followed by
    full sized loads.  */